MIPS IRIX compatibility vs. mips-elf

MIPS IRIX compatibility vs. mips-elf

[Daniel Jacobowitz]

Right now, a bunch of ld tests fail on mips-elf because .dynamic is not
first in the PT_DYNAMIC segment.  This happens because of a mismatch:
  - elf32_mips_irix_compat thinks that we are trying for Irix 5
    compatibility.  This causes _bfd_mips_elf_modify_segment_map
    to include a lot of other sections in .dynamic.
  - elf32bmip.sh thinks that this configuration is not compatible
    with Irix 5, so does not set TEXT_DYNAMIC.

As a result the .dynamic section is off in the writable segment,
causing PT_DYNAMIC to start at .dynsym, pass through .text, and reach
.dynamic (and then include an empty .data at the end, just for kicks).
The linker complains; the result is useless.

Of course, it's not quite clear what the value of mips-elf dynamic
executables is.  But it would be mildly nice if it worked, since that
gives us better test coverage for the -shared tests.  I'd like to turn
off the IRIX compat hacks for mips-elf - seems odd to include them. 
Alternately, we can just dodge the problem by disabling these tests for
mips*-elf.

Any opinions?
 
-- 
Daniel Jacobowitz

Re: MIPS IRIX compatibility vs. mips-elf

[Ian Lance Taylor]

Daniel Jacobowitz <drow@false.org> writes:

> Of course, it's not quite clear what the value of mips-elf dynamic
> executables is.  But it would be mildly nice if it worked, since that
> gives us better test coverage for the -shared tests.  I'd like to turn
> off the IRIX compat hacks for mips-elf - seems odd to include them. 
> Alternately, we can just dodge the problem by disabling these tests for
> mips*-elf.

I agree that we don't need Irix compatibility for the mips-elf target.

Ian

Re: MIPS IRIX compatibility vs. mips-elf

[sjhill]

> I agree that we don't need Irix compatibility for the mips-elf target.
> 
Sounds good to me. I'm only chiming in because I am working on IRIX
binary emulation on Linux.

-Steve

Re: MIPS IRIX compatibility vs. mips-elf

[Richard Sandiford]

Daniel Jacobowitz <drow@false.org> writes:
> Of course, it's not quite clear what the value of mips-elf dynamic
> executables is.  But it would be mildly nice if it worked, since that
> gives us better test coverage for the -shared tests.

Hmm.  Does it really make sense to test -shared on a mips*-elf target?
It's only really designed to handle static executables, so there's no
well-defined "right" output.

(Not that I'm objecting to removing IRIX compatibility, btw.)

Richard

Re: MIPS IRIX compatibility vs. mips-elf

[Daniel Jacobowitz]

On Fri, Jan 28, 2005 at 05:37:31PM +0000, Richard Sandiford wrote:
> Daniel Jacobowitz <drow@false.org> writes:
> > Of course, it's not quite clear what the value of mips-elf dynamic
> > executables is.  But it would be mildly nice if it worked, since that
> > gives us better test coverage for the -shared tests.
> 
> Hmm.  Does it really make sense to test -shared on a mips*-elf target?
> It's only really designed to handle static executables, so there's no
> well-defined "right" output.
> 
> (Not that I'm objecting to removing IRIX compatibility, btw.)

It's true that it's not clear what the right output is.  My precedent
is that it turned out assorted embedded folks actually do use shared
libraries with arm-elf (in one way or another).

Either's fine with me...

-- 
Daniel Jacobowitz

Re: MIPS IRIX compatibility vs. mips-elf

[Eric Christopher]

> > Hmm.  Does it really make sense to test -shared on a mips*-elf target?
> > It's only really designed to handle static executables, so there's no
> > well-defined "right" output.
> > 
> > (Not that I'm objecting to removing IRIX compatibility, btw.)
> 
> It's true that it's not clear what the right output is.  My precedent
> is that it turned out assorted embedded folks actually do use shared
> libraries with arm-elf (in one way or another).
> 
> Either's fine with me...
> 

I've seen embedded users use shared libraries as well. Not that I've
ever understood why but...

Anyhow, I'm good with removing the irix compatibility. Go ahead.

-eric